Skip to content
RoboticImagingPublic

About

Code release of FactorGP

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

FactorGP: A conditional-independence-aware approximate Gaussian process model for mapping

Overview

This is research code for the approximate GP model FactorGP, submitted to the RSS From Perception to Action workshop. Our model learns a GP approximating GMRF prior that can encode spatially varying lengthscales and conditional independences, capturing non-stationarity and discontinuities in data respectively.

The model code is contained within fgp/:

  • fgp/graph_learning.py: SVGD-based graph learning module
  • fgp/gbp.py : factor graph representation of the GMRF prior

Requirements

  • Python ≥ 3.9
  • Other dependencies:
pip install -r requirements.txt

Reproducing results

Terrain mapping (SRTM)

python active_mapping_srtm.py --env <environment_name> [--samples N]

Valid environments: N17E073, N43W080, N45W123, N47W124

Example:

python active_mapping_srtm.py --env N17E073
python active_mapping_srtm.py --env N43W080 --samples 700

Results are saved as results_<env>_seed<N>.png in the working directory and the eval table is printed to terminal.

Gas diffusion mapping

python active_sampling_gas.py

Per-step figures and final metrics are saved under figs/active_sampling/ and timestamped run data under results/active_sampling/. Eval table is printed to terminal.

Gas diffusion simulation

Code to produce gas fields via a forward-Euler Fick's law solver is in data/gas/diffusion_test.py. You can define your own building floor plan and source/sink locations to produce new simulated gas fields.

About

Code release of FactorGP

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages